Arnold Kling  

Models and Hunches

Will Obama Be a Carter or a Cl... Best Paragraph of the Day...

I am possibly going to speak on a panel in a few weeks on the issue of U.S. unemployment. I am thinking of trying to convey two ideas (is that too many? I only have ten minutes).

1. Do not trust anyone who answers the question, "How many jobs will (or did) policy X create?" If economic models were as reliable as decision-makers wanted them to be, we would not be where we are today.

2. Policy going forward will necessarily be based on hunches. Keynes' hunch was that there was too much saving. This was part of his general antipathy toward Victorian values. My hunch is that the recovery depends on profits. This is part of my general antipathy toward firms that raise funds from capital markets rather than from sales to customers (I had a front-row seat during the Internet bubble, when going public without profits was the standard business model).

I will elaborate a bit below.

1. Models and Delusions

Early in my career, I worked on econometric forecasting models at the Federal Reserve Board. I became very disillusioned with the reliability of those models. I still do not believe that we can use econometrics to answer important questions in macroeconomics. I should point out that the models that claim to estimate that the stimulus helped to increase employment are the same models that forecast that by now unemployment would be under 7.5 percent an falling.

The next step in my career was to work at Freddie Mac on models to forecast house price defaults. I think that those models were useful, but decision-makers did not want to hear how uncertain we were about our results. I remember when my boss and his managers had our first meeting with the Chief Operating Officer, and the COO asked us about the accuracy of our estimates of likely default costs in multifamily.

At the time, I was young and naive, so I blurted out an honest answer. I said, "factor of 2."

"Do you mean 2 basis points?" asked the COO. That would have meant that if we said expected default costs were 50 basis points, the right answer was between 48 and 52 basis points.

"No," I replied. "I mean that if we estimate 50 basis points, the answer could be anywhere from 25 basis points to 100 basis points." The COO became so incensed at my seeming flippancy that when we left we thought that he was going to fire all of us by the next day. Fortunately, that did not happen.

Years later, I can see that communication about the uncertainty in model estimates was not so blunt. If it had been, then nobody would have been convinced that you could create so much AAA paper out of subprime mortgages, or that regulators could allow the kind of leverage that they did at banks, which presumed narrow margin of error in estimates of mortgage defaults.

2. Hunches and Policies

John Maynard Keynes was part of a generation that rebelled against Victorianism. Keynes thought that the Victorian focus on saving and deferred gratification was excessive and outmoded. One can regard his macroeconomic recommendations as influenced by this visceral antipathy toward saving.

My own prejudice is that I dislike firms that raise funds in capital markets. I prefer firms that make profits and reinvest internally. My macroeconomic recommendations reflect this prejudice. I would like to focus on increasing profits and the demand for labor.

1. Eliminate the corporate income tax.
2. Eliminate all housing and mortgage subsidies, including the mortgage interest deduction.
3. Make all compensation taxable, including health insurance.
4. Eliminate the employer contribution to the payroll tax. This might be phased back in as the economy returns to full employment.

I am not sure how these four policies balance out in terms of the Budget. If they would increase the deficit, then I would try to find other spending programs to cut. If they would reduce the deficit, then I would not try to find offsetting tax cuts or spending increases.

Comments and Sharing

COMMENTS (4 to date)
Daniel Kuehn writes:

As general observations I think they're good, but I think you're going to get strong push-back on both from these angles:

On the first one, it's generally a good caution to have, but there's a difference between not trusting people who make those statements and not trusting people who make those statements and clearly acknowledge the uncertainty and caveat their assumptions.

Of course those are tough statements to make, but some people are very good and up front about the conditionality with which they provide their estimates. If you try to lump those responsible people who would be willing to discuss alternative guesses with people who simply claim a number for political purposes, you're going to (justifiably) get some resistance. However, given the example that you provide I think you're cognizant of this concern.

On your second point, I think you're just going to get resistance from people if you try to tell them that it was Keynes's anti-Victorianism that justified his concerns about excess savings (NOT savings in general, but excessive savings). You act as if the General Theory grew out of a counter-cultural impulse. I think you know that's not true, and even attempting to suggest it is going to get some resistance. Is it important to keep our biases in mind? Yes. But that doesn't mean it necessarily distorts our analysis. After all - presumably our analysis plays a major role in informing our biases, right! Do you propose libertarian solutions because you happen to have a libertarian philosophical disposition, or do you have a libertarian philosophical disposition because you've been repeatedly convinced by libertarian solutions?

Vacslav Glukhov writes:

The difference between 2 bp (or the expectation of) and the factor of two is not subtle and certainly is not quantitative. We can’t say whether the “true” default costs are 50 or 100 bps not because we don’t know, but because the “true” default probability does not exist.

This reminds me of Buchanan/Tullock’s approach: “there is no objective social truth waiting to be discovered”. There is no objective default rate waiting to be discovered. A model that assumes “objective truth” is doomed to fail.

Jeremy, Alabama writes:

Your points hang together as "#1, therefore #2". The variance of models is so wide that hunches provide a better basis for action.

Economic models DO have surprisingly large errors for the simplest things:

"When predicting the price of a commodity as simple as a carton of eggs five years into the future, there is a standard error of 15 percent."

Tom Grey writes:

Yes, models are far more uncertain than policy makers want -- both economic and climate.

Your policies to increase profits, then employment, seem politically unfeasible.
1. While dropping the corp income tax would be great (and many huge corps don't pay much, anyway), it's too obvious that this means higher taxes on people.
2. Ending house subsidies? Now??? This should have been pushed in 2003 (The Economist predicted housing boom bubble busting around then). Not now! It's a) a tax increase, b) a housing asset reduction / increase in the housing problem.

Consider an alternative: replace deduction for interest with flat 30% tax credit on the whole payment, both interest and principal BUT with a lifetime maximum credit of the 10 prior median taxpayer taxable incomes. (Last year median, not mean, was around $32000, so 10* is $320 000).
Lifetime maximum slowly increases with taxable wages, but total credit claimed will, after some 5 to 10 years, be reduced. At some point, people will have maxed out their claims.

3. Tax increase; but a good idea.
4. Tax decrease for business, but increasing the insolvency in SS. How could it be discussed without more comprehensive SS reform? Yet right now, an excellent stimulus for those who pay wages.

Comments for this entry have been closed
Return to top